Detection of PMSM Demagnetization Fault and Eccentricity Fault Based on Acoustic Images and DeiT Classifier

The acoustic signal contains a wealth of features and information, which can be used as a promising substitute for vibration signals in some cost-limited applications for motor fault diagnosis. This work presented a method involving the detection of motor rotor faults using acoustic signals, to diagnose the demagnetization faults and the eccentricity faults. The acoustic signals in different motor conditions are first de-noised by wavelet packet transform to remove the irrelevant components, then the MFCCs are extracted as the features by the reason of satisfactory performance in sound recognition. Next, the features are converted into 2D images, and the fault diagnosis and classification are realized by the DeiT classifier. Experiments show that compared with other time-frequency domain analysis, e.g., HHT, or the processing method that does not convert to 2D feature images, the accuracy of the proposed method is the highest, reaching 99.26%, proving that the method is reliable and effective for accurate fault diagnosis in a non-invasive manner.